mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

model: support GLM4V vision encoder (#18042)

Browse Source 
* convert ok

* no deepstack

* less new tensors

* cgraph ok

* add mrope for text model

* faster patch merger

* add GGML_ROPE_TYPE_MRNORM

* add support for metal

* move glm4v do dedicated graph

* convert: add norm_embd

* clip: add debugging fn

* working correctly

* fix style

* use bicubic

* fix mrope metal

* improve cpu

* convert to neox ordering on conversion

* revert backend changes

* force stop if using old weight

* support moe variant

* fix conversion

* fix convert (2)

* Update tools/mtmd/clip-graph.h

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* process mrope_section on TextModel base class

* resolve conflict merge

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Xuan-Son Nguyen
2025-12-16 11:25:26 +01:00
committed by GitHub
parent 9963b81f63
commit 3d86c6c2b5
17 changed files with 412 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gguf-py/gguf/constants.py
+7
View File
@@ -643,6 +643,7 @@ class MODEL_TENSOR(IntEnum):
V_MMPROJ_PEG = auto()
V_ENC_EMBD_CLS = auto()
V_ENC_EMBD_PATCH = auto()
V_ENC_EMBD_NORM = auto()
V_ENC_EMBD_POS = auto()
V_ENC_INPUT_NORM = auto()
V_ENC_ATTN_QKV = auto()
@@ -661,6 +662,7 @@ class MODEL_TENSOR(IntEnum):
V_LAYER_SCALE_2 = auto()
V_PRE_NORM = auto()
V_POST_NORM = auto()
V_MM_POST_NORM = auto()
V_MM_INP_NORM = auto()
V_MM_INP_PROJ = auto() # gemma3
V_MM_SOFT_EMB_NORM = auto() # gemma3
@@ -1016,6 +1018,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.V_MMPROJ_PEG: "mm.model.peg.{bid}",
MODEL_TENSOR.V_ENC_EMBD_CLS: "v.class_embd",
MODEL_TENSOR.V_ENC_EMBD_PATCH: "v.patch_embd",
MODEL_TENSOR.V_ENC_EMBD_NORM: "v.norm_embd",
MODEL_TENSOR.V_ENC_EMBD_POS: "v.position_embd",
MODEL_TENSOR.V_ENC_ATTN_QKV: "v.blk.{bid}.attn_qkv",
MODEL_TENSOR.V_ENC_ATTN_Q: "v.blk.{bid}.attn_q",
@@ -1034,6 +1037,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.V_LAYER_SCALE_2: "v.blk.{bid}.ls2",
MODEL_TENSOR.V_PRE_NORM: "v.pre_ln",
MODEL_TENSOR.V_POST_NORM: "v.post_ln",
MODEL_TENSOR.V_MM_POST_NORM: "mm.post_norm",
MODEL_TENSOR.V_MM_INP_PROJ: "mm.input_projection",
MODEL_TENSOR.V_MM_INP_NORM: "mm.input_norm",
MODEL_TENSOR.V_MM_SOFT_EMB_NORM: "mm.soft_emb_norm",
@@ -1094,6 +1098,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.V_MMPROJ_PEG,
MODEL_TENSOR.V_ENC_EMBD_CLS,
MODEL_TENSOR.V_ENC_EMBD_PATCH,
MODEL_TENSOR.V_ENC_EMBD_NORM,
MODEL_TENSOR.V_ENC_EMBD_POS,
MODEL_TENSOR.V_ENC_INPUT_NORM,
MODEL_TENSOR.V_ENC_ATTN_QKV,
@@ -1112,6 +1117,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.V_LAYER_SCALE_2,
MODEL_TENSOR.V_PRE_NORM,
MODEL_TENSOR.V_POST_NORM,
MODEL_TENSOR.V_MM_POST_NORM,
MODEL_TENSOR.V_MM_INP_PROJ,
MODEL_TENSOR.V_MM_INP_NORM,
MODEL_TENSOR.V_MM_SOFT_EMB_NORM,
@@ -3357,6 +3363,7 @@ class VisionProjectorType:
LIGHTONOCR = "lightonocr"
COGVLM = "cogvlm"
JANUS_PRO = "janus_pro"
GLM4V = "glm4v"
# Items here are (block size, type size)